American Economic Review 2014, 104(3): 963–990
http://dx.doi.org/10.1257/aer.104.3.963
When the Levee Breaks:
Black Migration and Economic Development
in the American South†
By Richard Hornbeck and Suresh Naidu*
In the American South, postbellum economic development may have
been restricted in part by white landowners’ access to low-wage
black labor. This paper examines the impact of the Great Mississippi
Flood of 1927 on black out-migration and subsequent agricultural
development. Flooded counties experienced an immediate and persistent out-migration of black population. Over time, landowners in
flooded counties modernized agricultural production and increased
its capital intensity relative to landowners in nearby similar nonflooded counties. Landowners resisted black out-migration, however,
benefiting from the status quo system of labor-intensive agricultural
production. (JEL J15, J43, N32, N52, N92, Q54, R23)
Underdeveloped societies often have a large population of low-wage agricultural
workers. Economic growth requires a sectoral reallocation of labor, yet various factors may keep workers in rural agriculture (Lewis 1954; Kuznets 1955; Brenner
1986; Banerjee and Newman 1998). Low-wage agricultural labor may discourage
labor-saving technological innovation (Habakkuk 1962; Allen 2009; Acemoglu
2010) or the adoption of new capital-intensive technologies (Atkinson and Stiglitz
1969; Basu and Weil 1998).
At the beginning of the twentieth century, Southern white planters dominated
areas with concentrated black populations. The Mississippi Delta exemplified this
system of racial inequality and discrimination that fostered paternalistic black labor
relations and narrowed black economic opportunities. The Great Mississippi Flood
of 1927 displaced workers and disrupted the traditional racial labor market equilibrium, leading to an exodus of black laborers and sharecroppers from flooded areas.
* Hornbeck: Littauer Center 232, Harvard University, Cambridge, MA 02138 (e-mail: [email protected].
edu); Naidu: 1405 IAB MC 3328, 420 W. 118th St., New York, NY 10027 (e-mail: [email protected]). For
many helpful comments and suggestions, we thank the referees, Ran Abramitzky, Daron Acemoglu, Lee Alston,
Hoyt Bleakley, Leah Boustan, Bill Collins, Paul David, Dave Donaldson, Price Fishback, Claudia Goldin, Larry
Katz, Paul Rhode, Andrei Shleifer, Jeff Vincent, Gavin Wright, and seminar participants at AEA, Arizona, BREAD,
Columbia, CIFAR, Harvard, Michigan, MIT, NBER, Pittsburgh, Stanford, UC Berkeley, and UC San Diego. For
financial research support, we thank CIFAR and the WCFIA’s Project on Justice, Welfare, and Economics. Tom
Beckford, James Feigenbaum, Lillian Fine, Andrew Das Sarma, and Leo Schwartz provided excellent research
assistance. The authors declare that they have no relevant or material financial interests that relate to the research
described in this paper.
† Go to http://dx.doi.org/10.1257/aer.104.3.963 to visit the article page for additional materials and author
disclosure statement(s).
963
964
THE AMERICAN ECONOMIC REVIEW
march 2014
This article examines the impact of the 1927 Mississippi flood on black
migration and agricultural development, emphasizing a link between black outmigration and the subsequent development of flooded areas. Empirical estimates
support historical accounts of a black exodus from flooded areas. Over time, agriculture became substantially more capital intensive and modernized in flooded
counties relative to nearby similar nonflooded counties. This induced agricultural
development was not associated with increased agricultural land values, however,
which is consistent with white landowners’ coercive efforts to resist black outmigration after the flood and maintain the status quo system of labor-intensive
agricultural production.
Using county-level data from the Censuses of Agriculture and Population, from
1900 to 1970, the empirical specifications compare changes between flooded
counties and nonflooded counties within the same state and with similar pre-1927
outcome values. Estimated increases in capital intensity are smaller in magnitude after controlling for differences in geography and 1930s New Deal program
spending, but flooded counties continue to have large relative increases over time
in the value of capital equipment and machinery. The analysis of black population declines is supplemented with individual-level census data, matched between
1920 and 1930.
In a similar analysis of counties near other major Southern rivers, compared to
counties further from rivers, there is little estimated relative change in black population or agricultural development in the absence of a flood. Counties near other major
Southern rivers exhibit some of the same outcome patterns prior to 1927, yet do not
experience the subsequent large relative changes estimated in flooded counties.
Our main interpretation of the flood’s impacts is that flood-induced black outmigration pushed flooded counties toward higher capital-intensity and larger-scale
farm operation, consistent with contemporary and historical qualitative accounts.
The empirical estimates appear less consistent with alternative interpretations, such
as direct impacts of the flood on capital or land. Further, general equilibrium impacts
on nonflooded counties appear to be small.
Following this temporary natural disaster, the persistent decline in black population is consistent with a breakdown of historically segmented labor markets for
black workers in the flooded region. While other factors may have also contributed
to later economic convergence throughout the US South, the aftermath of the 1927
Mississippi flood illustrates the potential for out-migration to spur economic development. In underdeveloped areas maintaining substantial populations of low-wage
agricultural workers, rural out-migration and decreased agricultural labor availability appears to have the potential to encourage subsequent agricultural development.
I. Historical Background
A. Southern Underdevelopment and the Mississippi Delta
In the US South, slavery and a geographic suitability for plantation agriculture
contributed to a system of labor-intensive agricultural production. Four million
slaves were emancipated during the Civil War; by 1900, however, most Southern
states had disenfranchised black populations (Naidu 2012). Southern white planters
VOL. 104 NO. 3
hornbeck and naidu: when the levee breaks
965
attempted to use their political influence to restrict black labor mobility.1 Planters
often pursued a strategy of paternalism to retain black workers, offering protection from white violence and implicit insurance. Protection was important for black
workers “because they lacked civil rights and society condoned violence” (Alston
and Ferrie 1999, p. 20). During a period of labor scarcity, a key factor for retaining
labor “was the landlord’s reputation among tenants with regard to his use of physical
violence” (Davis, Gardner, and Gardner 2009, p. 392).
The South remained underdeveloped between the Civil War and World War II,
relative to the North. While the North developed large manufacturing sectors, the
South remained primarily agricultural.2 Northern wheat threshing became increasingly mechanized in the nineteenth century (David 1975), while mechanization of
Southern cotton-picking was delayed until the mid-twentieth century.
Early cotton mechanization was mainly in planting and cultivation, where replacing mules and horses with tractors was associated with a 30 percent reduction in
labor inputs (Hurst 1933; Ellenberg 2007). Tractors and other labor-saving innovations were influential in American agricultural development (Olmstead and Rhode
2001; Gardner 2002; Steckel and White 2012), yet adoption lagged in the South.
High demand for harvest labor encouraged annual labor contracts and may have discouraged the partial mechanization of preharvest operations (Fleisig 1965; Whatley
1982, 1987): “not only cheap labor, but also the form of that cheap labor, reduced
the profitability of mechanization” (Whatley 1985, p. 1,208).
Wright (1986) describes a 1930 to 1970 economic transition from the “Old South”
to the “New South.” New technologies allowed full mechanization of cotton production, institutional changes contributed to a breakdown of sharecropping and regional
labor markets, and there was widespread black out-migration from rural agriculture.
Farm operation sizes increased as agriculture became more capital-intensive (Kirby
1987). Contemporaries recognized a feedback relationship between labor scarcity
encouraging agricultural mechanization and technological improvements displacing workers (Raper 1946). For later periods, some have emphasized the role of the
mechanical cotton picker in displacing workers (Day 1967; Grove and Heinicke
2003), while others have emphasized the impact of labor scarcity on mechanization
of the cotton harvest (Peterson and Kislev 1986; Holley 2000).
The Southern economy experienced remarkable growth in the mid-twentieth century. Much regional convergence in the United States was associated with increased
Southern agricultural wages and labor movement out of Southern agriculture
(Caselli and Coleman 2001). Potential contributing factors include advances in agricultural technology, the institutional breakdown of sharecropping, the New Deal,
World War II, malaria eradication, air conditioning, and Civil Rights regulation
(Arsenault 1984; Wright 1986; Heckman and Payner 1989; Donohue and Heckman
1991; Bleakley 2010; Besley, Persson, and Sturm 2010).
This paper focuses on the region surrounding the Mississippi-Yazoo Delta, which
has been dubbed the “most southern place on earth” (Cobb 1994). Powerful white
1 There has been substantial debate over these measures’ effectiveness and the degree of black labor mobility
(see, e.g., Myrdal 1944; Higgs 1973; Mandle 1978; Wright 1986; Fishback 1989; Margo 1990, 1991; Ransom and
Sutch 2001; Alston and Kauffman 2001; Naidu 2010).
2 In the US South, the share of employment in agriculture was more comparable to Southern and Eastern Europe.
966
THE AMERICAN ECONOMIC REVIEW
march 2014
planters recognized their economic dependence on local black labor, yet retaining
labor became increasingly tenuous through World War I and the first Great Migration
(1910–1930).
B. The Great Mississippi Flood of 1927
The Mississippi River historically changed course and spilled into natural floodplains. Over the late nineteenth and early twentieth centuries, levees were constructed to contain the river, and its natural spillways were closed. In 1926, the new
chief of the Army Corps of Engineers “for the first time officially stated in his annual
report that the levees were finally in condition ‘to prevent the destructive effect of
floods’” (Barry 1998, p. 175).
In 1927, the levee system failed catastrophically along the lower Mississippi River.
Heavy rains throughout the Mississippi River basin accumulated in rising river levels, and enormous pressure created 145 levee breaks that flooded 26,000 square
miles. In the three most affected states (Mississippi, Louisiana, Arkansas), flooding hit 36 percent of agricultural land and 29 percent of the population (Red Cross
1928). The flood caused $400 million in property damage and drowned 246 people.3
The Red Cross coordinated flood relief efforts, which focused on emergency shortterm needs (Red Cross 1928).4 The Red Cross established refugee camps that held
45 percent of the black population from flooded areas in Mississippi, Louisiana, and
Arkansas (Red Cross 1928).5 Refugee camp administration was placed under the
control of local counties and, in effect, powerful local white planters.
Southern planters faced the potential exodus of black workers, as the flood temporarily displaced workers and lowered labor demand for cotton harvesting in 1927.
Planters tried to retain their workers, turning many refugee camps into centers of
repression and racial abuse.6 Black work gangs were conscripted, and those caught
attempting to leave were beaten and returned.7 Flood aid was withheld unless blacks
worked on prescribed tasks, which planters justified on the grounds that rations
would “spoil” black workers and weaken the control planters had in “the old system” (Spencer 1994, p. 176). Amid stories of racial abuse, white planters in flooded
areas retained little credibility in offering paternalistic protection to their black
There was little flood insurance at this time (White 1945), and “[f ]looding in the Mississippi basin in 1927 and
1928 led the few companies that were selling cover to abandon the business” (Parker 2000, p. 413).
4 Of the $17 million spent, 30 percent was for food, and 14 percent was for livestock feed. The Red Cross spent
16 percent on seed for farmers to replant flooded cropland: two-thirds of this land could be replanted in 1927,
though the late planting season required some land to be shifted from cotton to corn, and the remaining lands were
replanted in 1928. Building construction, repairs, and household furnishings totaled 15 percent of expenditures, and
the remaining 25 percent was mainly for rescue and setup of refugee camps.
5 Refugee camps held 26 percent of the white population from flooded areas in these three states. The Red Cross
also gave relief outside of camps to 33 percent of the white population and 36 percent of the black population from
flooded areas.
6 Following directives from the Mississippi governor and the National Guard commander, the Red Cross issued
a memo on the “return of refugees,” stating: “Plantation owners desiring their labor to be returned from Refugee
Camps will make application to the nearest Red Cross representative,” whereupon they “will issue passes to refugees” (Barry 1998, pp. 313–314). The Delta and Pine Land Company, one of the nation’s largest cotton plantations,
established its own refugee camp and had its workers transferred by special train.
7 In May 1927, 21 black workers were caught and whipped by the National Guard for trying to escape a relief
camp (Spencer 1994, p. 177). In another case, a black insurance officer who refused to work was openly shot and
killed by the mayor of Lake Providence, Louisiana (Barry 1998, p. 330).
3 VOL. 104 NO. 3
hornbeck and naidu: when the levee breaks
967
workers.8 A circulated black newspaper, The Chicago Defender, described abuses
in Red Cross camps and listed job openings for blacks in Northern cities.
Despite planters’ efforts, or perhaps encouraged by such efforts, many black
families left flooded areas in search of better political and economic opportunities. Contemporary accounts describe black families leaving for Chicago and other
Northern cities. “The Afro-American reported that the relief camp experience had
‘inspired many backwoods farm[h]ands to make their first break for better things’”
(Spencer 1994, p. 177). Social networks shifted toward favoring migration; in
Greenville Mississippi, black leaders left for Chicago and crowds gathered at the
local railway station every Saturday night to see who was leaving and say goodbye
(Barry 1998).
Landowners’ accounts emphasize damages from losing their labor force, rather
than direct losses from the flood. One planter reported that “the most serious thing
that confronts the planter in the overflowed territory is the loss of labor, which is
great and is continuing” (Barry 1998, p. 416). Another planter lamented that “a
great deal of labor from the flooded section after being returned to the plantations
is going North … and it is going to offer a tremendous problem to all of us” (Barry
1998, p. 416).
White planters in flooded counties were forced to adapt to decreased availability
of black labor. In November 1927, the Engineering News Record noted: “In certain
sections of the lower Delta … where a crop could not be made this year two-thirds
to four-fifths of the families have moved away. In these districts farm-machinery
salesmen have been busy….” In 1931, a Mississippi Agricultural Extension bulletin discusses the “serious problem” of black out-migration and explores “the possible solution in mechanical farming,” comparing tenant-operated plantations and
tractor-operated plantations in the Delta (Vaiden, Smith, and Ayres 1931). Accounts
describe a reorganization of agricultural production and increased mechanization in
the Delta, even prior to the mechanical cotton picker: “Many planters have turned to
the use of wage labor and large-scale machinery in an effort to improve production
efficiency and decrease costs” (Langston and Thibodeaux 1939, p. 3).
The Mississippi Delta has often been examined as a microcosm of historical
Southern underdevelopment. After the 1927 flood, the Delta and its surrounding
region also provide a setting to explore economic development when particular
areas lose agricultural labor. Flooded areas lost black population due to the combined effects from temporary displacement and a decline in the opportunity cost of
migration, a breakdown of trust between planters and black workers, and a shift in
black social networks toward favoring migration.
8 One infamous Red Cross camp in the Delta was controlled by Will Percy, who forced blacks to work in the
camp free and wear laborer tags to receive food. “Following a killing of a black man by a white policeman on the
levees, Will Percy gave a condescending lecture to the black community at Mount Horeb church ‘Because of your
sinful, shameful laziness, because you refused to work on your own behalf unless you were paid, one of your race
has been killed.’ After this, the bond between the Percys and the blacks was broken” (Barry 1998, p. 333).
968
THE AMERICAN ECONOMIC REVIEW
march 2014
II. A Model of Flooding, Migration, and Agricultural Development
A. Model Setup
Assume that a representative Southern planter in county c and year t produces
agricultural goods for a world market with fixed prices: ​Ac​​ F(​Kct​ ​ , ​L​  Bct ​ ​  , ​L​  W
ct​  ​ ). Each
county has a fixed supply of land with productivity ​A​c​ . Capital ​K​ct​ is sufficiently
mobile or depreciable that the marginal return to capital r is equalized across counties. Labor is supplied inelastically by resident black workers ​L​  Bct ​​  and resident white
9
workers L
​ ​  W
ct​  ​.
Capital and labor are assumed to be substitutes, reflecting a choice between “Old
South” labor-intensive production and “New South” capital-intensive production.10
Capital is an important input in older production methods, but newer production
methods are embodied in capital goods. Black workers and white workers are also
substitutes, and we consider allowing for higher capital-labor substitutability for
black workers (e.g., due to differences in average education).11
White workers are perfectly mobile and earn a fixed outside “Northern” wage nor​ ​  B​or a home county wage​
malized to w
​ ​W​. Black workers can earn an outside wage w
B
w​  H ​ ​  . Planters have established a reputation for protecting their own workers from
racial violence, which is worth a to black workers in each period. Black workers also
pay a one-time moving cost M, equivalent to paying m in each future period, reflecting racially biased labor market institutions.12 As a consequence of black workers’
optimal migration decisions, home county wages are set in equilibrium such that​
w​  BH ​ ​  = ​w​  B​ − a − m. Each county is in an initial steady state with ​L​c 0​black workers.
In the first period, the Southern planter chooses inputs to maximize:
​   − r​Kc​1​ − (​w​  B​ − a − m)​L​  Bc1 ​ ​  − ​wW
​ ​ ​L​  W
​​  Bc1 ​ ​  ≤ ​L​  cB 0 ​.​  13
​Ac​​ F(​Kc​1​, ​L​  Bc1 ​,​  ​L​  W
c1 ​)
c1 ​,​  subject to L
B
B
We focus on the case in which this constraint binds and ​L​  c1 ​ ​  = ​L​  c 0 ​ ​  , consistent with
efforts by Southern planters to limit black out-migration. Capital investment and the
number of white workers are determined by
​  = r
(1) ​A​c​ ​FK​  ​ (​Kc​1​, ​L​  cB 0 ​,​  ​L​  W
c1 ​)
​  = ​wW
​ ​.
(2) ​A​c​ ​F​L​  W​ ​ ​(​Kc​1​, ​L​  cB 0 ​,​  ​L​  W
c1 ​)
9 We use agricultural “workers” to refer to wage laborers, share croppers, and share tenants who receive “wages”
in the form of cash, production shares, housing, and/or inputs.
10 In particular, we assume that the above production function represents an upper envelope over “Old South”
and “New South” technological choices. Increased use of “New South” methods is assumed to be “strongly black
labor-saving” (Acemoglu 2010); that is, in the case where machines replace black labor, the adoption of capitalintensive methods reduces the marginal product of black labor. Note, however, that output per black worker will still
increase following a decline in black labor availability and the adoption of black labor–saving technology.
11 In particular, we assume that: F is increasing and concave in all arguments; ∂
​​  2​F/∂ ​L​  Bct ​ ​  ∂ ​L​  W
ct​  ​ ≤ 0;
B
W
2
2
​∂​  ​F/∂ ​L​  ct ​∂
​  ​Kct
​ ​ < 0; and ∂
​ ​  ​F/∂ ​L​  ct​  ​∂ ​Kct
​ ​ ≥ 0. Note that the assumptions on the cross-partials rule out production
functions with constant returns to scale.
12 Anti-enticement laws made it illegal for one planter to hire another planter’s workers, while anti-vagrancy
laws made it illegal to be unemployed and without housing (Naidu 2010). The threat of force against black workers
shaped economic incentives and generated paternalist institutions, as in many other coercive labor markets (Naidu
and Yuchtman 2013; Acemoglu and Wolitzky 2011).
13 The planter could hire more black workers at wage w
​ ​  B​, but this would contradict the assumption of an initial
steady state with L
​ c ​ 0​black workers.
VOL. 104 NO. 3
hornbeck and naidu: when the levee breaks
969
In particular, equilibrium choices of capital and white workers depend on the initial
number of black workers: more black workers leads to a lower capital stock and
fewer white workers. Paternalism and moving costs both have the effect of lowering
planters’ labor costs in counties with more black workers.
B. Comparative Statics after the Flood
Consider the impact of a flood in some counties between periods 1 and 2. The
flood temporarily reduces the moving cost for black workers, either by imposing
some share of that cost or by reducing the opportunity cost of migration. Workers
are also housed in refugee camps controlled by the planter, and racial abuses in
these camps lower the planter’s ability to provide credible protection from white
violence. Black workers in refugee camps may also receive additional information
about Northern job opportunities or, as leaders of the black community migrate,
social networks may shift toward encouraging migration. The value of protection
falls to some fraction α, and the cost of moving falls to some fraction β, though the
planter may use a combination of incentives and threats to induce workers to return
at cost (1 − α)a + (1 − β)m.
After the flood, the Southern planter chooses inputs to maximize:
​   − r​Kc​2​ − (​wB​ ​ − αa − βm)​L​  Bc2 ​ ​  − ​wW
​ ​​L​  W
​ ​  Bc2 ​ ​  ≤ ​L​  cB 0 ​.​ 
​Ac​​ F(​Kc​2​, ​L​  Bc2 ​,​  ​L​  W
c2 ​)
c2 ​,​  subject to L
The flood effectively increases the cost of employing black workers. Assume that
the flood’s impacts are sufficiently large, i.e., α and β are sufficiently small, that
the constraint no longer binds, and the population of black workers declines in
equilibrium (​L​  Bc2 ​ ​  < ​L​  cB 0 ​)​  .
The loss of low-wage black workers increases planters’ labor costs, which encourages the adoption of labor-saving capital-intensive production methods. In flooded
counties, there will be increases in the capital stock, the population of white workers, and output per black worker. These changes will be especially pronounced
if there is a higher substitutability between capital and black workers; for example, if there is capital-skill complementarity and white workers are higher-skilled
on average.
This model does not include dynamic adjustment costs. It may take a number of
periods to make technological adjustments and to accumulate the desired capital
stock, particularly after destruction during the flood. Due to the temporary decline
in moving costs and permanent decline in paternalism value, however, the decline in
black population is predicted to be immediate and persistent.
Agricultural land values reflect the present discounted value of rents and, in this
baseline model, decline immediately due to the loss of exploitable low-cost black
labor. Land values would increase if capital investment became fixed to the land, but
this is a matter of accounting and does not reflect gains for landowners.
If there were sufficiently large externalities in capital investment, however, the
flood may cause a “big push” that increases land values immediately. The private
return to capital investment may be increasing in county-level total capital investment due to knowledge spillovers or coordinated investments in new capital equipment and infrastructure (see, e.g., Romer 1986; Murphy, Shleifer, and Vishny 1989;
Foster and Rosenzweig 1995). Agricultural modernization and capital investment
may also increase over time due to learning-by-doing, but land values would increase
970
THE AMERICAN ECONOMIC REVIEW
march 2014
immediately after the flood only if there were substantial externalities associated
with anticipated agricultural development.
III. Data Construction and Baseline Differences in Flooded Counties
A. Data Construction and Aggregate Trends
Historical county-level data are drawn from the Census of Agriculture and the
Census of Population (Haines 2010).14 The main variables of interest include: black
population, value of agricultural equipment and machinery, average farm size, number of tractors, number of mules and horses, and value of agricultural land and
buildings.15 The value of agricultural equipment and machinery includes all tools,
wagons, cotton gins, threshing machines, and all other machinery used in carrying out farm business (engines, motors, tractors, automobiles, and motor trucks);
note that this measure excludes the value of mules and horses, levees, or any land
improvements (Census Bureau 1927).
For the 1920s, a direct measure of migration is drawn from matched individuallevel census data in 1920 and 1930 (Boustan, Kahn, and Rhode 2012).16 The
match rate of 24 percent is comparable with the existing literature, though false
matches will tend to overstate migration rates. Later analysis examines the fraction
of matched individuals in 1930 that have left their 1920 county, state, or the South
(and differences by race).
The empirical analysis focuses on a balanced panel of 163 counties, from 1900
to 1970, for which data are available in every period of analysis. To account for
county border changes, data are adjusted in later periods to maintain 1900 county
definitions (Hornbeck 2010). The main sample is restricted to contiguous counties
in Arkansas, Louisiana, Mississippi, and Tennessee with a black population share
greater than 0.10 in 1920 and a fraction of cropland in cotton greater than 0.15 in
1920.17 Additional specifications examine counties elsewhere in the South, particularly those near other major rivers.
Figure 1 maps the extent of flooding in 1927, overlaid with county borders in
1900. Our main measure of flood intensity is the fraction of each county flooded
(from Figure 1), though the empirical results are robust to using Red Cross reports
on the fraction of population affected by flooding in each county.18
Figure 2 reports aggregate changes in the sample region from 1900 to 1970. Black
population decreased substantially from 1940 to 1970, during the second Great
Migration; and decreased somewhat in the 1910s, during the first Great Migration
(panel A). Total population increased through 1940, before declining into the 1960s
14 We thank Michael Haines and collaborators for providing additional data from ongoing collection.
Note that “farm size” refers to the size of farm operator parcels, rather than units of land ownership.
16 We thank Leah Boustan, Matt Kahn, and Paul Rhode for sharing their matched census data.
17 As an additional step in focusing the analysis on initially similar flooded and nonflooded counties, the empirical results are robust to controlling for counties’ estimated flood propensity score interacted with each year. The
probability that a county experienced any flooding is modeled as a probit function of the county’s black population
share in 1920 and fraction of cropland allocated to cotton in 1920. Only six of the original 163 counties are dropped
when limiting the sample to flooded and nonflooded counties with overlapping values of this propensity score.
18 Alternatively, the estimates are robust to using Red Cross data on the fraction of agricultural land flooded or
the fraction of total land flooded (Red Cross 1928). We thank Paul Rhode for sharing these Red Cross data, which
we supplemented.
15 VOL. 104 NO. 3
hornbeck and naidu: when the levee breaks
971
Sample county, 1900 boundary
Flooded sample region in 1927
Nonsample region (Arkansas, Louisiana,
Mississippi, Tennessee)
Figure 1. 1927 Flooded Region and Sample Counties (1900 Boundaries)
Notes: The 163 sample counties’ boundaries are based on county definitions in 1900. Countylevel data are adjusted to hold these boundaries fixed through 1970 (Minnesota Population
Center 2011). The sample region flooded in 1927 is shaded gray, based on a map compiled
and printed by the US Coast and Geodetic Survey (1927). The nonsample region is crosshatched. Excluded counties are missing outcome data in one of the analyzed years, have less
than 15 percent of reported cropland in cotton in 1920, or have a black population less than
10 percent of the total population in 1920.
(panel B). The value of agricultural capital increased through 1920 and then after
1940, during the second Great Migration, the introduction of the mechanical cotton
picker, and the Civil Rights movement (panel C). The number of mules and horses
were mainly constant through 1940, and then declined substantially through 1960
(panel D). Average farm sizes declined through 1930, before increasing substantially through 1970 (panel E). The value of agricultural land per farm acre increased
through World War I, and then again with post–World War II technological improvements (panel F). This figure provides some background on regional trends, whereas
the main empirical analysis estimates relative changes for flooded counties.
B. Baseline Differences in Flooded Counties
In an initial step, the empirical analysis explores predifferences between flooded
and nonflooded counties. For 1925 or 1920, depending on data availability, county
outcome Y is regressed on the fraction of county land flooded in 1927 and state fixed
effects:
(3) ​Y​ c​ = β Floo​dc​​ + ​αs​​ + ​ϵc​​ .
For each outcome variable, the estimated β reflects within-state differences in preflood characteristics for flooded counties and nonflooded counties.
972
THE AMERICAN ECONOMIC REVIEW
Panel A. log black population
Panel B. log population
9.4
10.3
9.3
10.2
march 2014
10.1
9.2
10
9.1
9.9
9
1900 1910 1920 1930 1940 1950 1960 1970
Panel C. log value of agricultural capital
1900 1910 1920 1930 1940 1950 1960 1970
Panel D. log number of mules and horses
16
9
15
8.5
14
8
13
7.5
12
7
1900 1910 1920 1930 1940 1950 1960 1970
Panel E. log average farm size
1900 1910 1920 1930 1940 1950 1960 1970
Panel F. Land value per farm acre
6
5.5
5
5
4
4.5
3
2
4
1900 1910 1920 1930 1940 1950 1960 1970
1900 1910 1920 1930 1940 1950 1960 1970
Figure 2. Aggregate Changes in the Sample Region (Arkansas, Louisiana, Mississippi, and Tennessee)
Note: Panels A–F report aggregated outcomes for the 163 sample counties in each year (Figure 1).
Source: Data are from the US Census of Agriculture and the US Census of Population.
To explore differences in pretrends between flooded and nonflooded counties,
equation (3) is modified to regress the change in outcome Y from 1910 to 1920
(or from 1920 to 1925) on the fraction of county land flooded in 1927 and state
fixed effects:
VOL. 104 NO. 3
973
hornbeck and naidu: when the levee breaks
Table 1—Baseline County Characteristics, by 1927 Flood Share
Panel A. Population in 1920
Black population share
Black population,
per 100 county acres
Population,
per 100 county acres
Black operated farm share
Panel B. Agriculture in 1925
Value of farm equipment,
per 100 county acres
Number of mules and horses,
per 100 county acres
Number of tractors
per 100 county acres
Average farm size
Farmland acres,
per 100 county acres
Value of farm land and buildings,
per 100 farm acres
Value of farm land and buildings,
per 100 county acres
Number of counties
log difference by 1927 flood share
Preflood levels
Preflood changes
Preflood
sample mean
(1)
Within-state
(2)
Controls
(3)
0.46
(0.20)
2.99
(2.46)
6.24
(4.33)
0.48
(0.26)
0.782**
(0.101)
1.003**
(0.171)
0.220
(0.133)
1.168**
(0.171)
0.449**
(0.133)
0.526*
(0.211)
0.077
(0.176)
0.697**
(0.201)
−0.003
(0.022)
0.033
(0.064)
0.037
(0.057)
0.019
(0.026)
−0.055
(0.029)
−0.052
(0.082)
0.003
(0.071)
−0.039
(0.034)
95.0
(60.9)
1.56
(0.84)
0.008
(0.010)
66.9
(21.4)
47.4
(17.3)
3,370
(2,094)
1,606
(1,316)
0.554**
(0.139)
0.422**
(0.141)
1.139**
(0.284)
−0.618**
(0.094)
−0.144
(0.102)
1.018**
(0.124)
0.875**
(0.168)
0.250
(0.178)
0.080
(0.172)
0.479
(0.390)
−0.417**
(0.101)
−0.244
(0.127)
0.702**
(0.162)
0.459*
(0.197)
−0.129
(0.079)
−0.080*
(0.040)
0.044
(0.115)
−0.048
(0.057)
0.017
(0.050)
−0.077
(0.045)
−0.272**
(0.046)
−0.350**
(0.061)
−0.076
(0.065)
−0.135*
(0.060)
−0.065
(0.060)
−0.200*
(0.081)
163
163
163
163
163
Within-state
(4)
Controls
(5)
Notes: Column 1 reports average baseline county characteristics in 1920 (panel A) and 1925 (panel B). All variables are reported in levels (not logs), and the standard deviation is reported in parentheses. Column 2 reports the
within-state difference for each county characteristic (in logs) by the fraction of the county flooded in 1927: the
coefficients are estimated by regressing the indicated county characteristic on the fraction of the county flooded in
1927 and a state fixed effect, weighting by county size. Column 3 reports the estimated difference when controlling
also for each county’s distance to the Mississippi River, geographic suitability for cotton and corn, terrain ruggedness, and longitude and latitude. Column 4 reports the within-state difference in pretrends for each county characteristic (in logs): panel A reports the change from 1910 to 1920, and panel B reports the change from 1920 to 1925.
The coefficients are estimated by regressing the change in the indicated county characteristic on the fraction of the
county flooded in 1927 and a state fixed effect, weighting by county size. Column 5 reports the estimated difference
in pretrends when controlling also for the above six county-level variables. Tractor data are only available in 162
counties and not before 1925. Robust standard errors are reported in parentheses.
** Significant at the 1 percent level.
* Significant at the 5 percent level.
(4) ​Y​ ct​ − ​Yc(t−1)
​ 
​ = β Floo​dc​​ + ​αs​​ + ​ϵ​c​ .
For each outcome variable, the estimated β reflects within-state differences in preflood trends in characteristics for flooded counties and nonflooded counties.
Table 1, column 1, reports average county characteristics prior to the 1927 flood.
Column 2 reports within-state differences in preflood characteristics for flooded
counties, and column 3 reports these differences conditional on six county-level
controls (distance from the Mississippi River, geographic suitability for cotton and
corn, terrain ruggedness, and longitude and latitude). Column 4 reports within-state
974
THE AMERICAN ECONOMIC REVIEW
march 2014
differences in preflood trends for flooded counties, and column 5 reports differential
trends conditional on the six county-level controls.
Prior to the 1927 flood, flooded counties and nonflooded counties are estimated
to have had similar changes in most outcomes. Flooded counties had an initially
higher black population and a greater intensity of small-scale agricultural production, though these differences are partly mitigated by county-level controls. To the
extent that flooded counties were different in pretrends or levels, the empirical specifications control for preflood differences.
IV. Empirical Framework
The empirical specifications estimate year-specific differences between flooded
counties and nonflooded counties, relative to a base year of 1925 or 1920. Outcome
Y in county c and year t is regressed on the fraction of county land flooded in 1927,
state-by-year fixed effects, and county fixed effects:
(5) ​Y​ ct​ = ​βt​​ Floo​d​c​ + ​αs​t​ + ​αc​​ + ​ϵ​ct​ .
Note that β is allowed to vary by year, so each estimated β is interpreted as the
average difference between flooded counties and nonflooded counties in that year
relative to the omitted base year of 1925 or 1920.
In practice, most empirical specifications control for county characteristics (​X​c​)
that may predict differential changes between flooded and nonflooded counties:
(6) ​Y​ ct​ = ​βt​​ Floo​dc​​ + ​αs​t​ + ​α​c​ + ​θt​​ ​Xc​​ + ​ϵ​ct​ .
The specifications control for preflood values of the outcome variable, flexibly
allowing for convergence over time in the outcome variable or otherwise differential
changes associated with initially different values.19 The identification assumption
is that, if not for the flood, flooded counties would have changed similarly to nonflooded counties in the same state and with similar preflood values of the outcome
variable. An empirical concern is that inherent differences between flooded and nonflooded areas may have caused some county characteristics to change differently
after 1927, even in the absence of the flood.20
As a consequence, most specifications also control for year-interacted measures
of counties’ distance to the Mississippi River, geographic suitability for cotton and
corn, terrain ruggedness, and longitude and latitude. Controlling for distance to the
Mississippi allows for the impact of river proximity to change over time, as counties
closer to the Mississippi are more likely to be flooded in 1927 and nearby counties
have better river access to markets.21 Controlling for crop suitability allows for cropspecific changes in technology and prices, or changes that otherwise ­differentially
19 Note that this specification is not a lagged dependent variable model; instead, the specification controls only
for pretreatment values of the dependent variable.
20 In particular, the assumption that flooded and nonflooded areas would have changed similarly becomes stronger in later periods.
21 Alternatively, the estimates are robust to restricting the sample to counties within 50 km or 100 km of the
Mississippi River.
VOL. 104 NO. 3
hornbeck and naidu: when the levee breaks
975
affect areas suitable for different crops.22 Controlling for terrain ruggedness allows
for differential changes in areas that may differ in suitability for agricultural mechanization.23 Finally, controlling separately for longitude and latitude allows for spatial patterns in economic changes that may be correlated with flooding.
The above control variables are fixed county geographic characteristics, but additional specifications also control for locally targeted economic policies. New Deal
programs in the 1930s may have impacted agricultural development by displacing tenants and increasing availability of harvest laborers (Whatley 1983; Depew,
Fishback, and Rhode 2012). New Deal spending by county may itself be affected by
the flood, but results are shown with and without year-interacted controls for counties’ per capita spending through the AAA, public works, relief, loan, and guaranteed mortgage programs (Fishback, Horrace, and Kantor 2005). The New Deal is a
sufficiently large shock to Southern agriculture, with potentially many of the same
effects as the flood, that these controls are a useful robustness check despite the
potential endogeneity of New Deal spending.
For the statistical inference in all specifications, standard errors are clustered at
the county level to adjust for heteroskedasticity and within-county correlation over
time. When allowing for spatial correlation among sample counties, the estimated
standard errors generally increase by less than 15 percent.24 The regressions are
weighted by county size, so the estimates reflect changes for an average acre of
flooded land.
V. Main Results
A. Population
Figure 3 shows estimated changes in black population for flooded counties, relative to changes for nonflooded counties, from estimating equation (5). Consistent
with the identification assumption, the black population share changed similarly
in flooded counties and nonflooded counties prior to the 1920s. Flooded counties
experienced a 14 percent (0.151 log point) decline from 1920 to 1930 in their black
population share. Following the 1927 flood, this short-run decline in black population share persisted through 1970.
Table 2, column 1, reports estimated declines in black population share when
estimating equation (6) and controlling for counties’ black population share in
1920, 1910, and 1900, and the six county geographic characteristics. Column 2
22 Cotton and corn are the two major crops in 1925 in the sample region. Crop suitability reflects the maximum
potential yield of that crop, as calculated by the FAO using data on climate, soil type, and ideal growing conditions
for that crop. The FAO’s Global Agro-Ecological Zone maps (version 3.0) are used to create county-level average
crop suitability for cotton and corn (FAO 2012). Potential yields are calculated using climate averages from 1961
to 1990 and rain-fed conditions with intermediate inputs.
23 Counties’ ruggedness is measured as the standard deviation in altitude across county points, calculated from
the USGS National Elevation Dataset (Farr et al. 2007). Estimates are similar when ruggedness is measured by
the maximum range in altitude across county points. Estimates are also similar when controlling for interactions
between terrain ruggedness and geographic suitability for cotton and corn (and their main effects).
24 Spatial correlation among counties is assumed to be declining linearly up to a distance cut off and zero after
that cut off (Conley 1999). For distance cut offs of 50 miles, 100 miles, or 200 miles, the estimated Conley standard
errors are generally less than 15 percent higher than the standard errors when clustering at the county level, depending on the outcome variable and year.
976
march 2014
THE AMERICAN ECONOMIC REVIEW
0.2
0.1
0
−0.1
−0.2
−0.3
−0.4
1900
1910
1920
1930
1940
1950
1960
1970
Figure 3. Estimated Differences in Black Population in Flooded Counties,
Relative to 1920
Notes: This graph reports estimated differences in log black population share between flooded
counties and nonflooded, relative to differences in 1920. From estimating equation (5) in the
text, the outcome is regressed on the fraction of the county flooded, state-by-year fixed effects,
and county fixed effects. The dashed lines indicate 95 percent confidence intervals, based on
robust standard errors clustered by county.
Table 2—Estimated Differences in Population by Flood Share, Relative to 1920
Decade
1930
1940
1950
1960
1970
Counties
log fraction black
(1)
(2)
−0.144** −0.155**
(0.043)
(0.041)
−0.165** −0.207**
(0.047)
(0.050)
−0.202** −0.240**
(0.065)
(0.065)
−0.170* −0.221**
(0.080)
(0.070)
−0.146
−0.207*
(0.096)
(0.088)
163
163
log black population
(3)
(4)
log population
(5)
(6)
−0.170*
(0.078)
−0.107
(0.085)
−0.181**
(0.065)
−0.168*
(0.081)
−0.008
(0.070)
0.061
(0.075)
−0.032
(0.059)
0.023
(0.062)
163
163
163
163
−0.218
(0.118)
−0.277
(0.145)
−0.344*
(0.170)
−0.226*
(0.093)
−0.222*
(0.104)
−0.202
(0.115)
−0.006
(0.108)
−0.063
(0.149)
−0.207
(0.178)
−0.010
(0.085)
−0.005
(0.116)
−0.024
(0.144)
log black farm share
(7)
(8)
−0.254** −0.239**
(0.058)
(0.057)
−0.245** −0.259**
(0.068)
(0.074)
−0.309** −0.345**
(0.098)
(0.105)
163
163
Notes: Each column reports estimated changes in the indicated outcome variable: changes in flooded counties relative to changes in nonflooded counties, relative to the omitted year of 1920. Columns 1, 3, 5, and 7 report coefficients from regressing the outcome variable on the fraction of the county flooded in 1927, state-by-year fixed
effects, county fixed effects, and lagged values of the outcome variable in 1900, 1910, and 1920 interacted with
each year, as well as controlling for six county geographic characteristics interacted with each year (distance to the
Mississippi River, cotton and corn suitability, ruggedness, and latitude and longitude). Columns 2, 4, 6, and 8 add
five categories of New Deal spending interacted with year as controls. All regressions are weighted by county size.
Robust standard errors clustered by county are reported in parentheses.
** Significant at the 1 percent level.
* Significant at the 5 percent level.
VOL. 104 NO. 3
hornbeck and naidu: when the levee breaks
977
reports estimated declines when also controlling for counties’ New Deal spending
in the 1930s.
The demographic shift was mainly caused by a decline in the black population
(Table 2, columns 3 and 4), with little change in total population (Table 2, columns 5 and 6). Theoretical predictions for changes in total population depend on
the functional form of the production function, but the offsetting increase in white
population is consistent with a complementarity between capital and white labor
and/or an inelastic demand for labor.25
These changes in population are reflected in the composition of farms, where
there was a large decline in the share of black-operated farms (Table 2, columns 7
and 8). Black farm operators tended to be lower on the “tenancy ladder” than white
farm operators, with much higher rates of sharecropping.26 This black population
may have the highest moving costs and the most to gain from employer paternalism.
The decrease in the supply of black workers and farm operators, and a shift toward
white agricultural labor and farm operators, would be associated with increased
labor costs for landowners.
The estimated changes in county-level population mainly reflect net migration,
but a direct measure of out-migration uses matched individual-level census data
from 1920 and 1930.27 Table 3, panel A and column 1, reports that the fraction
of matched people leaving their county between 1920 and 1930 is 11.8 percentage points higher in flooded counties than in nonflooded counties within the same
state.28 Flooded counties also have a higher fraction of matched people leaving their
state (column 2), though a similar fraction leaving the South entirely (column 3).29
Migration estimates are more striking for the subsample of individuals whose race
is observed. Blacks in flooded counties are more likely to leave their county (column 1), their state (column 2), and the South entirely (column 3), while whites are
not more likely to leave flooded counties. Column 4 reports estimated differences
in migration into flooded counties, with an insignificant and negative effect overall
(panel A), little effect on blacks (panel B), and higher in-migration among whites
(panel C). Analysis of migration rates is based on smaller samples of matched individuals, however, so the estimates are less precise and more sensitive to county-level
controls than estimated changes in county-level population.30
Overall, the estimates are consistent with historical accounts of an immediate
and persistent decline in black population in flooded counties. The empirical results
do not identify whether this decline in population reflects the flood’s temporary
Increases in farmland (estimated below) imply subsequent declines in total population per farm acre.
Using data from Jefferson County, Arkansas (Alston and Ferrie 2005) for the period from 1920 to 1927, we
calculate that 66 percent of white-operated farms are owner-operated and 74 percent of black-operated farms are
either sharecroppers (45 percent) or cash-tenancies (29 percent).
27 We thank Leah Boustan, Matt Kahn, and Paul Rhode for sharing their matched census data (Bouston, Kahn,
and Rhode 2012). Matches are required to be unique by name and place of birth (state or country) within five-year
age bands. Average migration rates may be overstated due to false matches, though this would introduce bias only
if errors were differential across flooded counties and nonflooded counties.
28 The county-level out-migration rate is calculated as the number of matched people leaving the county, divided
by the total number of matched people originally in the county. The regression is weighted by the number of
matched people in each county, so the estimates reflect a change in probability of migration for the average person.
29 Southern states are defined as Arkansas, Louisiana, Tennessee, Mississippi, Alabama, North Carolina, South
Carolina, Georgia, and Florida.
30 Some of the estimates lose statistical significance when controlling for county geographic differences, but the
estimates continue to indicate higher out-migration among blacks than whites.
25 26 978
march 2014
THE AMERICAN ECONOMIC REVIEW
Table 3—Estimated Differences in 1920–1930 Migration Rates by Flood Share
Panel A. All matched people
Mean in nonflooded counties
Difference in flooded counties
Number of matched individuals
Counties with matched people
Panel B. Black population only
Mean in nonflooded counties
Difference in flooded counties
Number of matched individuals
Counties with matched people
Panel C. White population only
Mean in nonflooded counties
Difference in flooded counties
Number of matched individuals
Counties with matched people
Fraction moving
out of county
(1)
Fraction moving
out of state
(2)
Fraction moving
out of South
(3)
Fraction moving
into county
(4)
0.661
(0.130)
0.118**
(0.031)
7,102
162
0.287
(0.136)
0.113**
(0.033)
7,102
162
0.161
(0.123)
−0.003
(0.040)
7,102
162
0.768
(0.469)
−0.268
(0.158)
6,822
162
0.698
(0.206)
0.139**
(0.046)
1,186
153
0.245
(0.187)
0.177**
(0.041)
1,186
153
0.136
(0.158)
0.068*
(0.030)
1,186
153
0.585
(0.544)
0.007
(0.125)
917
153
0.629
(0.152)
−0.069
(0.058)
1,324
156
0.296
(0.183)
0.034
(0.060)
1,324
156
0.166
(0.162)
−0.032
(0.050)
1,324
156
0.374
(0.330)
0.336**
(0.123)
1,124
156
Notes: Column 1 reports the fraction of people, matched between the 1920 census and 1930 census, that left their
county between 1920 and 1930. Column 2 reports the fraction of matched people that have left their state from 1920
to 1930, and Column 3 reports the fraction that have left the South from 1920 to 1930. Column 4 reports the number
of people moving into that county between 1920 and 1930, as a fraction of people in that county in 1920. Panel A
includes all matched people, panel B limits the sample to those people known to be black, and panel C limits the
sample to those people known to be white. Each panel and column reports the mean value in nonflooded counties and the standard deviation in parentheses. Each panel and column reports the difference in migration rate for
flooded counties, relative to non-flooded counties, controlling for state fixed effects. All regressions are weighted by
the fraction of matched people in each county. Robust standard errors are reported in parentheses.
** Significant at the 1 percent level.
* Significant at the 5 percent level.
d­ isplacement effect and a decline in the opportunity cost of migration, a breakdown
of trust between planters and black workers, or a shift in black social networks
toward favoring migration. The persistent decline in black population, following
a temporary natural disaster, does suggest a breakdown of historically segmented
labor markets for black workers in the flooded region. The empirical analysis following explores the impacts on subsequent agricultural development.
B. Agricultural Capital Investment and Modernization
Figure 4 shows estimated changes in the value of agricultural capital equipment
and machinery for flooded counties, relative to nonflooded counties, from estimating equation (5). The value of capital had been changing similarly in flooded and
nonflooded counties and, following losses sustained during the 1927 flood, mostly
recovered by 1930. By 1940, the value of agricultural capital had increased substantially in flooded counties relative to nonflooded counties. Relative increases in
VOL. 104 NO. 3
979
hornbeck and naidu: when the levee breaks
1.5
1.2
0.9
0.6
0.3
0
−0.3
1900
1910
1920
1930
1940
1950
1960
1970
Figure 4. Estimated Differences in Farm Capital in Flooded Counties,
Relative to 1920
Notes: This graph reports estimated differences in log value of farm equipment and machinery between flooded counties and nonflooded, relative to differences in 1925. From estimating equation (5) in the text, the outcome is regressed on the fraction of the county flooded,
state-by-year fixed effects, and county fixed effects. The dashed lines indicate 95 percent confidence intervals, based on robust standard errors clustered by county.
agricultural capital continued through 1970. While it becomes more difficult to
rule out other differential shocks in later periods, capital values do not appear to
converge over time in flooded and nonflooded counties. The estimated increases in
capital intensity are smaller in magnitude after controlling for differences in geography (Table 4, column 1) and also for 1930s New Deal program spending (Table 4,
column 2), but flooded counties continue to have large relative increases over time
in the value of capital equipment and machinery.
Table 4, columns 3 and 4, report large percent increases in tractors in flooded
counties relative to nonflooded counties through at least the 1940s.31 Tractors were
still rare in the sample region during the 1920s and 1930s, however, so these estimates reflect only small initial increases in the number of tractors per farm.32
Traditional sources of agricultural power were mules and horses, which were
used by agricultural workers but were overall a substitute for manpower.33 Table 4,
columns 5 and 6, report that the number of mules and horses was relatively
unchanged. Despite the large number of animal deaths during the flood, there is no
indication of planters reducing use of this older-vintage capital in the immediate
31 While tractor quality is unobserved, higher agricultural capital in later periods and a more-similar number of
tractors may indicate higher tractor quality in flooded counties.
32 Based on the average number of tractors per farm in nonflooded counties in each year, the estimates in column 3 represent an increase in the number of tractors per farm of 0.01 in 1930, 0.04 in 1940, 0.06 in 1945, 0.17 in
1954, and 0.62 in 1970.
33 Mules and horses are a form of “capital,” but their value is not included in the value of agricultural capital
equipment and machinery.
980
march 2014
THE AMERICAN ECONOMIC REVIEW
Table 4—Estimated Differences in Capital Intensity by Flood Share, Relative to 1925
log farm capital
Decade
1930
1935
1940
1945
(1)
−0.070
(0.108)
(2)
−0.101
(0.109)
0.378** 0.326**
(0.113)
(0.119)
1950
1954
1960
log tractors
(3)
0.473*
(0.193)
(4)
0.505*
(0.202)
0.951** 0.910**
(0.261) (0.258)
0.622** 0.530**
(0.204) (0.185)
0.403
(0.209)
0.311
(0.181)
0.455*
(0.204)
0.262
(0.177)
162
162
1964
1970
Counties
0.807** 0.673**
(0.159)
(0.160)
163
163
log mules and horses
(5)
0.048
(0.058)
0.104
(0.066)
0.155*
(0.076)
(6)
−0.042
(0.056)
0.014
(0.065)
0.050
(0.079)
−0.242
(0.130)
−0.460**
(0.138)
−0.278
(0.142)
−0.468**
(0.159)
163
163
log average farm size
(7)
0.037
(0.059)
0.196**
(0.063)
0.185*
(0.083)
0.284**
(0.078)
0.484**
(0.092)
0.609**
(0.108)
0.782**
(0.145)
0.942**
(0.167)
0.723**
(0.162)
163
(8)
0.042
(0.064)
0.187**
(0.062)
0.171*
(0.083)
0.242**
(0.077)
0.416**
(0.091)
0.536**
(0.107)
0.719**
(0.142)
0.891**
(0.166)
0.644**
(0.162)
163
Notes: Each column reports estimated changes in the indicated outcome variable: changes in flooded counties relative to changes in nonflooded counties, relative to the omitted year of 1925. Columns 1, 3, 5, and 7 report coefficients from regressing the outcome variable on the fraction of the county flooded in 1927, state-by-year fixed
effects, county fixed effects, and lagged values of the outcome variable in 1900, 1910, and 1920 interacted with
each year, as well as controlling for six county geographic characteristics interacted with each year (distance to the
Mississippi River, cotton and corn suitability, ruggedness, and latitude and longitude). Columns 2, 4, 6, and 8 add
five categories of New Deal spending interacted with year as controls. All regressions are weighted by county size.
Robust standard errors clustered by county are reported in parentheses.
** Significant at the 1 percent level.
* Significant at the 5 percent level.
aftermath of the flood. By the 1950s and 1960s, however, use of this “Old South”
power source declined.34
Many agricultural production practices are unobserved in the data, but increased
scale of farm operation was strongly associated with a transition from older methods of agricultural production to modernized agricultural production in the South.
Table 4, columns 7 and 8, report that flooded counties experienced a gradual and
substantial increase in average farm size, relative to nonflooded counties. Farm sizes
increased particularly during the 1950s and 1960s as mechanical cotton pickers
became increasingly available.35
It is difficult to measure the increase in labor productivity associated with reported
changes in production inputs and methods. As a proxy, however, data are available
for the value of crops per capita. From estimating equation (6), the log value of
crops per capita changed similarly in flooded and nonflooded counties from 1910
through 1930. This proxy for average labor productivity increased substantially in
34 Note that it is more difficult to rule out other differential shocks in these later periods.
Grove and Heinicke (2003) present aggregate data showing that the share of cotton harvested by mechanical
pickers in the Mississippi Delta region went from essentially zero in 1949 to 60 percent by the 1960s.
35 VOL. 104 NO. 3
hornbeck and naidu: when the levee breaks
981
flooded counties through the 1930s, 1940s, and 1950s.36 While county-level wage
data are unavailable, the estimated increases in labor productivity are consistent
with increased labor costs in flooded counties.
Overall, the estimated increases in agricultural capital equipment and machinery
appear to embody labor-saving technological adoption in the agricultural sector.
Initial increases in agricultural capital did not come through replacing older-vintage
capital that was destroyed by the flood (i.e., mules and horses). While many agricultural production practices are unobserved, the observed trend toward larger farm
operation size would be associated with a bundle of modernized agricultural production methods.
The flooded area was likely too small, relative to national cotton production, for the
decline in black labor availability to encourage labor-saving technological innovation
(e.g., Habakkuk 1962; Allen 2009; Acemoglu 2010), but decreased labor availability appears to have made flooded counties more suitable for ­capital-intensive methods of production (e.g., Atkinson and Stiglitz 1969; Basu and Weil 1998). Increases
over time in the value of capital and farm sizes may reflect ­learning-by-doing,
increased availability of wage workers during the Depression and New Deal, and/or
subsequent increases in the availability of mechanical cotton pickers.
C. Farmland Acreage and Value
Table 5, columns 1 and 2, report that flooded counties began to experience
substantial relative increases in farmland in the 1940s and 1950s. Thus, as farms
became larger and more capital intensive, agricultural production in flooded counties also became more land intensive. One interpretation is that increased capital
usage was complementary with clearing and plowing additional farmland.37 The
estimated increases in farmland are smaller in magnitude when controlling for
1930s New Deal program spending, but there remains a substantial relative trend
toward increased farmland in flooded counties.
Estimated increases in total farmland complicate an analysis of the value of agricultural land and buildings. In principle, changes in agricultural land values reflect
the loss (or gain) to landowners from decreased black labor availability and subsequent agricultural adaptation. New farmland may be of generally lower quality than
initial farmland, however, causing a downward bias in the value of farmland per
farm acre. By contrast, clearing and plowing new farmland requires some sunk costs
and, as these costs are capitalized into land values, there will be an upward bias in
the value of farmland per county acre.
Immediately after the flood, flooded counties experienced little change or
declines in the value of agricultural land per farm acre (Table 5, columns 3 and 4)
36 When excluding the New Deal controls, the estimated coefficients (and standard errors) are 0.034 (0.131)
in 1930, 0.254 (0.123) in 1940, 0.575 (0.244) in 1950, and 1.272 (0.310) in 1960. When including the New Deal
controls, the estimated coefficients are −0.014 (0.135) in 1930, 0.214 (0.123) in 1940, 0.468 (0.242) in 1950, and
1.078 (0.299) in 1960. Estimated relative changes are similar and slightly larger for the log value of crops per person
living in rural areas of the county.
37 The increase in agricultural land may represent a decrease in land under the public domain, or an increase in
the fraction of privately owned land that is in operation (or fallow) and captured by census enumerators. Note that
the empirical specifications estimate relative changes, so the reported increases may also reflect less of a decline in
farmland in some flooded counties relative to nonflooded counties.
982
march 2014
THE AMERICAN ECONOMIC REVIEW
Table 5—Estimated Differences in Farmland by Flood Share, Relative to 1925
log farmland
Decade
1930
1935
1940
1945
1950
1954
1960
1964
1970
Counties
(1)
(2)
log value of farmland
per farm acre
(3)
(4)
log value of farmland
per county acre
(5)
(6)
−0.023
(0.047)
0.053
(0.058)
0.203**
(0.059)
0.265**
(0.080)
0.325**
(0.086)
0.408**
(0.101)
0.558**
(0.129)
0.669**
(0.144)
0.943**
(0.180)
−0.101
(0.053)
−0.009
(0.059)
0.096
(0.069)
0.176*
(0.085)
0.188*
(0.091)
0.284**
(0.105)
0.377**
(0.134)
0.485**
(0.147)
0.701**
(0.176)
−0.043
(0.065)
−0.084
(0.069)
−0.110
(0.056)
−0.333**
(0.078)
−0.391**
(0.080)
−0.351**
(0.078)
−0.404**
(0.108)
−0.266**
(0.099)
−0.301**
(0.081)
−0.007
(0.062)
−0.073
(0.067)
−0.091
(0.055)
−0.313**
(0.079)
−0.295**
(0.070)
−0.235**
(0.070)
−0.295**
(0.106)
−0.208*
(0.101)
−0.223**
(0.081)
−0.119
(0.066)
−0.074
(0.093)
0.055
(0.074)
−0.046
(0.089)
−0.133
(0.100)
−0.042
(0.102)
0.013
(0.129)
0.216
(0.125)
0.401*
(0.155)
−0.128*
(0.061)
−0.080
(0.087)
0.033
(0.075)
−0.058
(0.097)
−0.128
(0.097)
−0.007
(0.100)
0.041
(0.131)
0.213
(0.136)
0.368*
(0.152)
163
163
163
163
163
163
Notes: Each column reports estimated changes in the indicated outcome variable: changes in flooded counties relative to changes in nonflooded counties, relative to the omitted year of 1925. Columns 1, 3, and 5 report coefficients
from regressing the outcome variable on the fraction of the county flooded in 1927, state-by-year fixed effects,
county fixed effects, and lagged values of the outcome variable in 1900, 1910, and 1920 interacted with each year,
as well as controlling for six county geographic characteristics interacted with each year (distance to the Mississippi
River, cotton and corn suitability, ruggedness, and latitude and longitude). Columns 2, 4, and 6 add five categories
of New Deal spending interacted with year as controls. All regressions are weighted by county size. Robust standard
errors clustered by county are reported in parentheses.
** Significant at the 1 percent level.
* Significant at the 5 percent level.
and per county acre (Table 5, columns 5 and 6). The value of land per farm acre
declined further over time, which may reflect a compositional decline in average
land quality. The value of land per county acre increased over time, which may
reflect sunk costs in clearing and plowing new farmland.38 Across all four specifications, however, the estimates reject a substantial immediate increase in agricultural
land values that might suggest landowners anticipated benefiting from the forced
economic transition.39
Overall, the estimates suggest a single equilibrium in which landowners must
adapt to decreased labor availability. Landowners’ coordinated resistance to black
out-migration is consistent with landowners not anticipating economic gains from a
“big push” toward agricultural modernization. Black migrants presumably benefited
38 Landowners in flooded counties may also have unexpectedly benefited from later technological innovations
that favored capital-intensive agricultural production.
39 Data on land values and building values are available separately, by decade, from 1900 to 1940. In 1920, the
value of land averages 77 percent of the combined value of land and buildings. Focusing on changes in the value of
land only, in 1930 and 1940, the estimates are similar.
VOL. 104 NO. 3
hornbeck and naidu: when the levee breaks
983
from the option to migrate after the flood, though this does not imply that migrants
benefited overall from the flood.40
VI. Threats to Validity
A. Falsification Exercise
An empirical concern is that nonflooded areas may be an inherently poor control
group for flooded areas, despite including controls for preflood outcomes and geographic characteristics. As an alternative check on the results, a falsification exercise explores whether there are also differential changes between counties close to
other major Southern rivers and counties further from other major Southern rivers.
Restricting the analysis to nonflooded states, this sample includes 171 counties
within 50 km of a major river and 72 counties between 50 km and 150 km of a
major river.41
As in the main sample, counties near other major Southern rivers have higher
black population in 1920 and a greater intensity of small-scale agricultural production in 1925 than counties further from other major Southern rivers. As in the main
sample, these counties had been experiencing similar trends in the county outcomes
of interest.42
Table 6 reports that counties near other major Southern rivers changed similarly
after 1927 to counties further from other major Southern rivers.43 Of the few statistically significant estimates, counties close to other major Southern rivers experienced some relative declines in capital.
Overall, in the absence of a catastrophic flood, counties near other major Southern
rivers do not experience the black out-migration and increased agricultural development that appeared in counties flooded by the Mississippi in 1927. While the
Mississippi River is a special river within the Southern United States, other counties near major rivers showed many of the same differences in characteristics prior
to 1927. These estimates lend support to the identification assumption that flooded
counties would have changed similarly to nonflooded counties in the absence of the
flood.
40 Under free mobility, there would need to be some externality or coordination failure among migrants to generate welfare gains from flood-induced out-migration when the black population had previously chosen to stay in the
region. Chay and Munshi (2012) examine Southern black migration networks in the early twentieth century, which
are consistent with potential externalities.
41 These cut offs reflect typical distances to the Mississippi for flooded counties and nonflooded counties, respectively. As in the main sample, the sample is restricted to counties with a black population share greater than 0.10 in
1920 and a fraction of cropland in cotton greater than 0.15 in 1920. The major rivers shapefile was obtained from
ESRI (2002).
42 Analogous to the estimates from Table 1, these estimates refer to modified versions of equations (3) and (4),
where the fraction of county flooded is replaced with a dummy variable for whether the county is within 50 km of
a major river.
43 In a modified version of equation (6), the fraction of county flooded is replaced with a dummy variable for
whether the county is within 50 km of a major river. The specification controls for changes over each time period
that are correlated with state and initial outcome differences. We have omitted three columns due to space constraints (population, mules and horses, value of farmland per county acre), but there are no substantial or statistically significant changes in these outcomes.
984
march 2014
THE AMERICAN ECONOMIC REVIEW
Table 6—Estimated Relative Changes in Counties within 50 km of Other Major Southern Rivers
1930
1935
1940
1945
1950
1954
1960
1964
1970
Counties
Fraction
black
(1)
Black
Black
Value farm
population farm share
capital
(2)
(3)
(4)
Tractors
(5)
0.018
(0.027)
0.034
(0.021)
−0.045
(0.039)
−0.071
(0.067)
0.025
(0.017)
0.025
(0.033)
0.042
(0.030)
−0.092
(0.059)
0.016
(0.024)
0.007
(0.044)
0.012
(0.035)
−0.087
(0.088)
−0.109
(0.093)
−0.010
(0.031)
0.002
(0.061)
−0.016
(0.038)
0.019
(0.071)
243
243
0.009
(0.013)
243
Average
farm size
(6)
−0.190*
(0.091)
−0.254*
(0.119)
0.052*
(0.023)
0.014
(0.031)
0.041
(0.031)
0.065
(0.037)
0.055
(0.038)
0.043
(0.039)
0.056
(0.044)
0.078
(0.048)
0.091
(0.050)
243
240
243
−0.238*
(0.112)
Land value/
Farmland
farm ac
(7)
(8)
0.021
(0.016)
0.024
(0.018)
0.029
(0.020)
0.002
(0.022)
−0.008
(0.026)
−0.006
(0.031)
−0.011
(0.044)
−0.009
(0.048)
−0.031
(0.061)
−0.013
(0.027)
0.002
(0.037)
−0.015
(0.037)
−0.042
(0.041)
0.016
(0.042)
−0.028
(0.048)
−0.042
(0.047)
−0.054
(0.046)
−0.020
(0.042)
243
243
Notes: Each column reports estimated changes in the indicated outcome variable (in logs): changes in counties
within 50 km of a major river relative to changes in counties within 50 km–150 km of a major river, relative to the
omitted year of 1920 or 1925. The sample is restricted to Southern counties within 150 km of a major river, excluding all counties in the main sample region (Figure 1). The indicated outcome variable is regressed on a dummy for
whether the county is within 50 km of a major river, state-by-year fixed effects, county fixed effects, and county
outcome values in 1900, 1910, 1920, and 1925 (when available), interacted with each year. All regressions are
weighted by county size. Robust standard errors clustered by county are reported in parentheses.
** Significant at the 1 percent level.
* Significant at the 5 percent level.
B. Plantation Counties
One particular empirical concern is that flooded counties are more likely to be
“plantation counties,” as recorded by Brannen (1924) for 1910.44 In nonflooded
Southern states, plantation counties are estimated to have little differential change
from 1930 to 1940 in the main outcome variables of interest.45 Between 1940 and
1970, however, plantation counties do experience some relative declines in black
population and increases in agricultural capital, average farm size, and farmland.46
44 A plantation is defined as a “unified agricultural organization of considerable size under one management, of
practically a continuous tract of land, operated as a single unit with respect to the methods of control of labor and
products, all of which may be worked by wage hands, or all or a part of which may be subdivided and let to tenants”
(Brannen 1924, p. 9). Brannen used since-lost census data and judgment to select counties where “plantation farming in these counties is known to be important” (Brannen 1924, p. 69).
45 In a modified version of equation (6), the fraction of county flooded is replaced with a dummy variable for
whether the county is a “plantation county.”
46 In a modified version of equation (6), as described above, plantation counties have some statistically insignificant relative declines in black population. Relative to 1925, agricultural capital is similar in plantation counties
through 1940 and higher by 0.23 log points by 1970.
VOL. 104 NO. 3
hornbeck and naidu: when the levee breaks
985
In the main sample of flooded and nonflooded counties, the empirical results
are robust to controlling for differential changes in plantation counties.47 Further,
the impacts of the flood on agricultural development are not driven by plantation
counties. Allowing for heterogeneous effects of the flood on plantation counties
and ­nonplantation counties, the nonplantation counties experience clear declines in
black population and increases in agricultural development.48
Related concerns are that flooded counties may have different preflood black
tenancy shares or manufacturing outcomes, which could contribute to differential
changes over the mid-twentieth century. The empirical estimates are robust to controlling for differential changes over time that are correlated with pre-1927 values
for the share of tenants who are black, the number of manufacturing establishments,
and average manufacturing wages.49
C. Alternative Interpretations of the Flood’s Impacts
Our main interpretation of the flood’s impacts is that flood-induced black outmigration pushed flooded counties toward higher capital intensity and larger-scale
farm operation, consistent with contemporary and historical qualitative accounts.
However, there are two other main channels through which the 1927 flood may have
had lasting economic impacts. First, the flood may have caused general economic
disruption and the replacement of vintage capital stocks with more technologically
advanced capital.50 Second, the flood may have changed land productivity.
In the first case, by causing general economic disruption, the flood may have
encouraged landowners to reevaluate and update agricultural production.51 In particular, reconstruction may have replaced destroyed “vintage” capital goods with
newer capital goods, leading to a short-run increase in capital investment and modernized capital equipment in flooded areas. As capital stocks depreciated in nondamaged areas, however, natural replacement would have led to convergence in the
quantity and age of capital goods.
The empirical results are generally inconsistent with this first alternative interpretation. The value of agricultural capital equipment and machinery is found to diverge
over time in flooded counties, rather than increase immediately and c­ onverge over
As in equation (6), the specification controls for a dummy variable for whether the county is a plantation
county (interacted with year).
48 In a modified version of equation (6), the fraction of county flooded is interacted with a dummy variable
for whether the county is a “plantation county” and a dummy variable for whether the county is a “nonplantation
county.”
49 In estimating equation (6) with pre-1927 outcome values and the six geographic controls, the additional
­year-interacted control variables include: the log fraction of tenants who are black (in 1900, 1910, and 1920); the
log number of manufacturing establishments (in 1900 and 1920); and log average manufacturing wages (in 1900
and 1920).
50 Related alternative explanations are that the flood could have encouraged the coordination and consolidation
of land holdings or induced a series of foreclosures that allowed new entrepreneurial farmers to enter. Land ownership was fairly concentrated and stable in this region, so we do not focus on these related alternative explanations.
To the extent that landowners attempted to coordinate investments and production, this coordination was mainly
in maintaining the status quo labor-intensive system rather than coordinating over land assembly and increased
mechanization.
51 The lower Mississippi region had an unfortunate history of natural disasters in the early twentieth century
(Boustan, Kahn, and Rhode 2012); while none were as large as the 1927 flood, this was a volatile region that
appears less likely to have settled into economic complacency.
47 986
THE AMERICAN ECONOMIC REVIEW
march 2014
time. There was also no immediate decline in older-vintage capital, such as mules
and horses.
Historically high levels of capital depreciation imply that post-flood capital reconstruction would have had few persistent “vintage capital” effects. While tractors are
among the more durable capital goods, an approximate annual depreciation rate
of 12 percent implies that roughly 85 percent of investment in 1927 would have
depreciated by 1935 (Hurst 1933). Investment in agricultural buildings may have
been more durable; from estimating equation (6), however, the value of agricultural
buildings in flooded counties declined slightly by 1930 and 1940.52
In the second case, by changing land productivity, the flood may have directly
impacted land values and factor demand. While repeated historical flooding of the
Mississippi contributed to the formation of productive soils, one isolated flood
would have had limited direct benefits for soil productivity. The flood also damaged
land improvements, but these were generally rebuilt quickly and new lands were
improved and brought under cultivation in flooded counties.53 It is difficult to know
whether the 1927 flood and the subsequent 1928 Flood Control Act increased or
decreased landowners’ expected flood risk, though there should be less differential
change in perceived future risk once controlling for distance to the Mississippi River
or limiting the sample to counties near the Mississippi River.54
From estimating equation (6), flooded counties experienced little immediate
change in cotton productivity or corn productivity.55 In subsequent years, cotton
and corn acreages expanded, and there was little systematic change in productivity.
These estimates are also consistent with literature on early mechanization being
labor saving but not yield increasing (Hayami and Ruttan 1985).
Finally, for interpreting the main results, the flood may have had general equilibrium impacts on nearby nonflooded counties. The empirical estimates overstate the
aggregate impact of the flood for particular outcomes that are affected oppositely in
nonflooded counties. Our interpretation of the results focuses mainly on the flood’s
relative impacts, however, such as changes in the relative availability of black labor
and the relative change in agricultural capital and scale of farm operation.
The flood may be expected to have had little indirect impact on nonflooded counties in subsequent years and decades, even if the flood initially disrupted nonflooded
counties. There may even have been small immediate impacts on nonflooded counties’ output prices and return on capital, given the degree of integration in agricultural
markets and the small share of agricultural output directly affected by the flood. As
a test of the magnitude of local economic spillovers, Table 7 reports the estimated
change in counties bordering the flooded region, relative to counties 100 km from
52 Data on land values and building values are available separately, by decade, from 1900 to 1940. The log value
of building values, per farm acre or per county acre, is regressed on the fraction of the county flooded in 1927,
state-by-year fixed effects, county fixed effects, and county outcome values in 1900, 1910, and 1920, interacted
with each year.
53 Red Cross efforts to introduce new varieties of crops and livestock were generally limited (Red Cross 1928),
as reconstruction efforts were focused on emergency needs and temporary relief.
54 The 1928 Act was mandated to protect all of the potentially flooded counties, not just those that were actually
flooded, and thus involved substantial upriver tributaries rather than a sole focus on levees. Further, reconstruction
and modification of the levee system had little direct effect on available agricultural land, irrigation, or drainage.
55 The log quantity of cotton or corn yielded per harvested acre is regressed on the fraction of the county flooded
in 1927, state-by-year fixed effects, county fixed effects, and county outcome values in 1900, 1910, 1920, and 1925,
interacted with each year (with and without controls for geography and New Deal spending).
VOL. 104 NO. 3
987
hornbeck and naidu: when the levee breaks
Table 7—Estimated Changes in Counties Bordering the Flooded Region,
Relative to Counties 100 km Away
1930
1935
1940
1945
1950
1954
1960
1964
1970
Counties
Fraction
black
(1)
Black
Black
Value farm
population farm share
capital
(2)
(3)
(4)
0.003
(0.048)
−0.025
(0.036)
0.010
(0.064)
Tractors
(5)
0.035
(0.034)
0.007
(0.057)
−0.026
(0.040)
−0.048
(0.066)
0.075
(0.052)
0.050
(0.081)
0.008
(0.050)
0.260
(0.218)
0.179
(0.158)
0.101
(0.064)
0.068
(0.107)
0.102
(0.076)
0.077
(0.127)
94
94
0.019
(0.023)
94
Average
farm size
(6)
−0.036
(0.084)
0.046
(0.121)
−0.022
(0.027)
−0.029
(0.030)
−0.060*
(0.030)
−0.035
(0.040)
−0.078
(0.053)
−0.112
(0.057)
−0.105
(0.064)
−0.070
(0.068)
−0.001
(0.064)
94
94
94
0.045
(0.140)
0.084
(0.117)
Land value/
Farmland
farm ac
(7)
(8)
−0.015
(0.028)
−0.014
(0.037)
−0.032
(0.042)
−0.007
(0.053)
−0.048
(0.056)
−0.081
(0.061)
−0.130
(0.073)
−0.116
(0.082)
−0.111
(0.111)
94
0.060
(0.039)
0.022
(0.050)
0.047
(0.046)
0.102*
(0.051)
0.129*
(0.056)
0.114*
(0.053)
0.076
(0.055)
0.151**
(0.053)
0.118**
(0.042)
94
Notes: Each column reports estimated changes in the indicated outcome variable: changes in counties bordering the
flooded region relative to changes in counties 100 km from the flooded region, relative to the omitted year of 1920
or 1925. The sample is restricted to the 94 main sample counties with no flooding (Figure 1). The indicated outcome variable is regressed on the (negative) distance from the flooded region in 100-km units, state-by-year fixed
effects, county fixed effects, and county outcome values in 1900, 1910, 1920, and 1925 (when available), interacted with each year. All regressions are weighted by county size. Robust standard errors clustered by county are
reported in parentheses.
** Significant at the 1 percent level.
* Significant at the 5 percent level.
the flood border.56 Consistent with small local economic spillovers, particularly in
the immediate aftermath of the flood, there was little change in counties bordering
the flooded region compared to further counties.
VII. Conclusion
The Great Mississippi Flood of 1927 was a transformative event for areas in and
around the Mississippi Delta. In a region infamous for oppressive racial institutions,
the flood led to an exodus of black agricultural workers. This persistent decline in
black population, following a temporary natural disaster, is consistent with a breakdown of historically segmented labor markets for black workers in the flooded region.
Each outcome variable is regressed on the (negative) distance from the flooded region in 100-km units, stateby-year fixed effects, county fixed effects, and county outcome values in 1900, 1910, 1920, and 1925 (when available), interacted with each year. An increase in distance from 0 km to 100 km is equivalent to an increase from the
closest counties to the eightieth centile. We have omitted three columns due to space constraints (population, mules
and horses, value of farmland per county acre), but there are no substantial or statistically significant changes in
these outcomes.
56 988
THE AMERICAN ECONOMIC REVIEW
march 2014
Unable to prevent a decline in black labor availability, planters in flooded counties
appear to have been pushed over time toward greater usage of agricultural capital
and a modernized scale of farm operation. Estimated increases in capital intensity
are smaller in magnitude after controlling for differences in geography and 1930s
New Deal program spending, but flooded counties continue to have large relative
increases over time in the value of capital equipment and machinery. This induced
agricultural development was not associated with increased agricultural land values,
however, which is consistent with white landowners’ coercive efforts to resist black
out-migration after the flood and maintain the status quo system of labor-intensive
agricultural production for their own benefit.
The aftermath of the 1927 Mississippi flood illustrates the potential for decreased
agricultural labor availability to spur agricultural development. In underdeveloped
areas maintaining substantial populations of low-wage agricultural workers, inducing rural out-migration and decreasing agricultural labor availability appears to have
the potential to encourage subsequent agricultural development.
REFERENCES
Acemoglu, Daron. 2010. “When Does Labor Scarcity Encourage Innovation?” Journal of Political
Economy 118 (6): 1037–78.
Acemoglu, Daron, and Alexander Wolitzky. 2011. “The Economics of Labor Coercion.” Econometrica
79 (2): 555–600.
Allen, Robert C. 2009. The British Industrial Revolution in Global Perspective, New Approaches to
Economic and Social History. Cambridge, UK: Cambridge University Press.
Alston, Lee J., and Joseph P. Ferrie. 1999. Southern Paternalism and the American Welfare State: Eco-
nomics, Politics, and Institutions in the South, 1865–1965. Cambridge, UK: Cambridge University
Press.
Alston, Lee J., and Joseph P. Ferrie. 2005. “Time on the Ladder: Career Mobility in Agriculture, 1890–
1938.” Journal of Economic History 65 (4): 1058–81.
Alston, Lee J., and Kyle D. Kauffman. 2001. “Competition and the Compensation of Sharecroppers
by Race: A View from Plantations in the Early Twentieth Century.” Explorations in Economic History 38 (1): 181–94.
Arsenault, Raymond. 1984. “The End of the Long Hot Summer: The Air Conditioner and Southern
Culture.” Journal of Southern History 50 (4): 597–628.
Atkinson, Anthony B., and Joseph E. Stiglitz. 1969. “A New View of Technological Change.” Economic Journal 79 (315): 573–78.
Banerjee, Abhijit, and Andrew F. Newman. 1998. “Information, the Dual Economy, and Development.” Review of Economic Studies 65 (4): 631–53.
Barry, John M. 1998. Rising Tide: The Great Mississippi Flood of 1927 and How It Changed America.
New York: Simon & Schuster.
Basu, Susanto, and David N. Weil. 1998. “Appropriate Technology and Growth.” Quarterly Journal of
Economics 113 (4): 1025–54.
Besley, Timothy, Torsten Persson, and Daniel M. Sturm. 2010. “Political Competition, Policy and
Growth: Theory and Evidence from the US.” Review of Economic Studies 77 (4): 1329–52.
Bleakley, Hoyt. 2010. “Malaria Eradication in the Americas: A Retrospective Analysis of Childhood
Exposure.” American Economic Journal: Applied Economics 2 (2): 1–45.
Boustan, Leah Platt, Matthew E. Kahn, and Paul W. Rhode. 2012. “Moving to Higher Ground: Migration Response to Natural Disasters in the Early Twentieth Century.” American Economic Review
102 (3): 238–44.
Brannen, Claude O. 1924. Relation of Lands Tenure to Plantation Organization. Washington, DC: US
Government Printing Office.
Brenner, Robert. 1986. “The Social Basis of Economic Development.” In Analytical Marxism, edited
by John Roemer, 23–53. Cambridge: University of Cambridge Press.
Caselli, Francesco, and Wilbur John Coleman. 2001. “The U.S. Structural Transformation and
Regional Convergence: A Reinterpretation.” Journal of Political Economy 109 (3): 584–616.
VOL. 104 NO. 3
hornbeck and naidu: when the levee breaks
989
Chay, Kenneth, and Kaivan Munshi. 2012. “Black Networks after Emancipation: Evidence from
Reconstruction and the Great Migration.” Unpublished.
Cobb, James C. 1994. The Most Southern Place on Earth: The Mississippi Delta and the Roots of
Regional Identity. New York: Oxford University Press.
Conley, T. G. 1999. “GMM Estimation with Cross Sectional Dependence.” Journal of Econometrics
92 (1): 1–45.
David, Paul A. 1975. Technical Choice Innovation and Economic Growth. Cambridge, UK: Cambridge
University Press.
Davis, Allison, Burleigh B. Gardner, and Mary R. Gardner. 2009. Deep South: A Social Anthropologi-
cal Study of Caste and Class. Columbia, SC: University of South Carolina Press.
Day, Richard H. 1967. “The Economics of Technological Change and the Demise of the Sharecrop-
per.” American Economic Review 57 (3): 427–49.
Depew, Briggs, Price Fishback, and Paul Rhode. 2012. “New Deal or No Deal in the Cotton South: The
Effect of the AAA on the Agriculture Labor Structure.” Unpublished.
Donohue, John J., and James Heckman. 1991. “Continuous versus Episodic Change: The Impact of Civil
Rights Policy on the Economic Status of Blacks.” Journal of Economic Literature 29 (4): 1603–43.
Ellenberg, George B. 2007. Mule South to Tractor South: Mules, Machines, Agriculture, and the Trans-
formation of the Cotton South. Tuscaloosa, AL: University of Alabama Press.
Environmental Systems Research Institute, Inc. (ESRI). 2002. “US Rivers (ESRI).” Redlands, CA.
Available at ftp://ftp.igsb.uiowa.edu/GIS_Library/USA/us_rivers.zip.
FAO. 2012. Global Agro-Ecological Zones (GAEZ v3.0). IIASA, Laxenburg, Austria and FAO, Rome.
Farr, T. G., et al. 2007. “The Shuttle Radar Topography Mission.” Rev. Geophys., 45, RG2004.
Fishback, Price V. 1989. “Debt Peonage in Postbellum Georgia.” Explorations in Economic History
26 (2): 219–36.
Fishback, Price V., William C. Horrace, and Shawn Kantor. 2005. “Did New Deal Grant Programs
Stimulate Local Economies: A Study of Federal Grants and Retail Sales during the Great Depression.” Journal of Economic History 65 (1): 36–71.
Fleisig, Heywood. 1965. “Mechanizing the Cotton Harvest in the Nineteenth-Century South.” Journal
of Economic History 25 (4): 704–06.
Foster, Andrew D., and Mark R. Rosenzweig. 1995. “Learning by Doing and Learning from Others:
Human Capital and Technical Change in Agriculture.” Journal of Political Economy 103 (6): 1176–
1209.
Gardner, Bruce L. 2002. American Agriculture in the Twentieth Century: How It Flourished and What
It Cost. Cambridge, MA: Harvard University Press.
Grove, Wayne A., and Craig Heinicke. 2003. “Better Opportunities or Worse? The Demise of Cotton
Harvest Labor, 1949–1964.” Journal of Economic History 63 (3): 736–67.
Habakkuk, H. J. 1962. American and British Technology in the Nineteenth Century: The Search for
Labour-Saving Inventions. New York: Cambridge University Press.
Haines, Michael R. 2010. “Historical, Demographic, Economic, and Social Data: The United States,
1790–2002.” ICPSR02896-v3. Ann Arbor, MI: Inter-university Consortium for Political and Social
Research [distributor]. doi:10.3886/ICPSR02896.v3.
Hayami, Yujiro, and Vernon W. Ruttan. 1985. Agricultural Development: An International Perspective. Baltimore, MD: Johns Hopkins Press.
Heckman, James J., and Brook S. Payner. 1989. “Determining the Impact of Federal Antidiscrimination Policy on the Economic Status of Blacks: A Study of South Carolina.” American Economic
Review 79 (1): 138–77.
Higgs, Robert. 1973. “Race, Tenure, and Resource Allocation in Southern Agriculture, 1910.” Journal
of Economic History 33 (1): 149–69.
Holley, Donald. 2000. The Second Great Emancipation: The Mechanical Cotton Picker, Black Migration, and How They Shaped the Modern South. Fayetteville, AR: University of Arkansas Press.
Hornbeck, Richard. 2010. “Barbed Wire: Property Rights and Agricultural Development.” Quarterly
Journal of Economics 125 (2): 767–810.
Hornbeck, Richard, and Suresh Naidu. 2014. “When the Levee Breaks: Black Migration and Economic Development in the American South: Dataset.” American Economic Review. http://dx.doi.
org/10.1257/aer.104.3.963.
Hurst, W. M. 1933. Power and Machinery in Agriculture. Washington, DC: US Department of Agriculture.
Kirby, Jack Temple. 1987. Rural Worlds Lost: The American South, 1920–1960. Baton Rouge: LSU
Press.
990
THE AMERICAN ECONOMIC REVIEW
march 2014
Kuznets, Simon. 1955. “Economic growth and income inequality.” American Economic Review 45
(1): 1–28.
Langston, Ethelbert Lee, and Ben Hur Thibodeaux. 1939. “Plantation Organization and Operation in
the Yazoo-Mississippi Delta Area.” Mississippi Ag Bulletin 682.
Lewis, W. Arthur. 1954. “Economic Development with Unlimited Supplies of Labour.” The Manches-
ter School 22 (2): 139–91.
Mandle, Jay R. 1978. The Roots of Black Poverty: The Southern Plantation Economy after the Civil
War. Durham, NC: Duke University Press.
Margo, Robert A. 1990. Race and Schooling in the South, 1880–1950: An Economic History. Chicago:
University of Chicago Press.
Margo, Robert A. 1991. “Segregated Schools and the Mobility Hypothesis: A Model of Local Govern-
ment Discrimination.” Quarterly Journal of Economics 106 (1): 61–73.
Minnesota Population Center. 2011. National Historical Geographic Information System: Version 2.0.
Minneapolis, MN: University of Minnesota.
Murphy, Kevin M., Andrei Shleifer, and Robert W. Vishny. 1989. “Industrialization and the Big Push.”
Journal of Political Economy 97 (5): 1003–26.
Myrdal, Gunnar. 1944. An American Dilemma: The Negro Problem and Modern Democracy. New
Brunswick, NJ: Transaction Publishers.
Naidu, Suresh. 2010. “Recruitment Restrictions and Labor Markets: Evidence from the Postbellum US
South.” Journal of Labor Economics 28 (2): 413–45.
Naidu, Suresh. 2012. “Suffrage, Schooling, and Sorting in the Post-Bellum US South.” National
Bureau of Economic Research Working Paper 18129.
Naidu, Suresh, and Noam Yuchtman. 2013. “Coercive Contract Enforcement: Law and the Labor Mar-
ket in Nineteenth Century Industrial Britain.” American Economic Review 103 (1): 107–44.
Olmstead, Alan L., and Paul W. Rhode. 2001. “Reshaping the Landscape: The Impact and Diffusion
of the Tractor in American Agriculture, 1910–1960.” Journal of Economic History 61 (3): 663–98.
Parker, Dennis J. 2000. Floods. Vol. 1. New York: Routledge.
Peterson, Willis, and Yoav Kislev. 1986. “The Cotton Harvester in Retrospect: Labor Displacement or
Replacement?” Journal of Economic History 46 (1): 199–216.
Ransom, Roger L., and Richard Sutch. 2001. One Kind of Freedom: The Economic Consequences of
Emancipation. Cambridge, UK: Cambridge University Press.
Raper, Arthur. 1946. “The Role of Agricultural Technology in Southern Social Change.” Social Forces
25 (1): 21–30.
Red Cross. 1928. “Reports and Statistics.” RG 200, Boxes 733–735, Folders 224.02, 224.031, and
224.08.
Romer, Paul M. 1986. “Increasing Returns and Long-run Growth.” Journal of Political Economy 94
(5): 1002–37.
Spencer, Robyn. 1994. “Contested Terrain: The Mississippi Flood of 1927 and the Struggle to Control
Black Labor.” Journal of Negro History 79 (2): 170–81.
Steckel, Richard H., and William J. White. 2012. “Engines of Growth: Farm Tractors and Twentieth-
Century US Economic Welfare.” National Bureau of Economic Research Working Paper 17879.
US Census Bureau. 1927. United States Census of Agriculture 1925. Washington, DC: US Govern-
ment Printing Office.
US Coast and Geodetic Survey. 1927. “Mississippi River Flood of 1927 [map]” Records of the
Coast and Geodetic Survey, RG 23. http://www.archives.gov/global-pages/larger-image.
html?i=/publications/prologue/2007/spring/images/coast-miss-flood-l.jpg&c=/publications/
prologue/2007/spring/images/coast-miss-flood.caption.html.
Vaiden, M. G., James Owen Smith, and W. E. Ayres. 1931. “Making Cotton Cheaper, Can Present Production Cost Be Reduced?” Mississippi Agricultural Bulletin 290 (S): 413–45.
Whatley, Warren C. 1982. Institutional Change and Mechanization in the Cotton South. PhD. diss.,
Stanford University.
Whatley, Warren C. 1983. “Labor for the Picking: The New Deal in the South.” Journal of Economic
History 43 (4): 905–29.
Whatley, Warren C. 1985. “A History of Mechanization in the Cotton South: The Institutional Hypothesis.” Quarterly Journal of Economics 100 (4): 1191–1215.
Whatley, Warren C. 1987. “Southern Agrarian Labor Contracts as Impediments to Cotton Mechanization.” Journal of Economic History 47 (1): 45–70.
White, Gilbert F. 1945. Human Adjustment to Floods: A Geographical Approach to the Flood Problem
in the United States. Chicago: University of Chicago Press.
Wright, Gavin. 1986. Old South, New South: Revolutions in the Southern Economy since the Civil War.
New York: Basic Books.